ida-domain-mcp
Headless MCP server for IDA Pro enabling on-demand database loading and programmatic reverse engineering workflows via MCP tools.
README
ida-domain-mcp
A headless Model Context Protocol (MCP) server for IDA Pro built on top of ida-domain and ida-pro-mcp. It lets AI agents (or any MCP client) open and analyze IDA databases on demand — without launching the IDA manully — and control common reverse engineering workflows programmatically.
Unlike GUI-centric approaches, ida-domain-mcp spins up per-project worker processes on demand and loads binaries via an MCP tool call during the agent's workflow. You don't have to pre-load binaries at MCP server startup, and once configured, the whole flow can run fully automatically without human interaction.
Example
<!--
-->

Why it's different
- Headless by design: No dependency on the IDA graphical UI. Uses IDA's headless runners underneath via
ida-domain. - On-demand database loading: Call the
open_databaseMCP tool at any time during the agent session to load a binary or IDB; no manual preloading required. - Multi-project isolation: Each
project_nameruns in its own worker process; multiple binaries can be analyzed concurrently without interfering with each other.
Features (tools)
High-level categories of tools exposed via MCP:
- Project/session management:
open_database,close_database - IDB/Core:
idb_meta,lookup_funcs,cursor_addr,cursor_func,list_funcs,list_globals,imports,strings,segments,local_types,int_convert - Analysis & Xrefs:
decompile,disasm,xrefs_to,xrefs_to_field,callees,callers,entrypoints,analyze_funcs,find_bytes,find_insns,basic_blocks,find_paths,search,find_insn_operands,export_funcs,callgraph,xref_matrix,analyze_strings - Debugger:
dbg_start,dbg_exit,dbg_continue,dbg_run_to,dbg_step_into,dbg_step_over,dbg_list_bps,dbg_add_bp,dbg_delete_bp,dbg_enable_bp,dbg_regs,dbg_regs_thread,dbg_regs_cur,dbg_gpregs_thread,dbg_current_gpregs,dbg_regs_for_thread,dbg_current_regs,dbg_callstack,dbg_read_mem,dbg_write_mem - Memory & Patch:
get_bytes,get_u8,get_u16,get_u32,get_u64,get_string,get_global_value,patch - Types & Structs:
declare_type,apply_types,infer_types,structs,struct_info,read_struct,search_structs,stack_frame,declare_stack,delete_stack - Python:
py_eval
Most MCP tools here are thin wrappers around upstream ida_pro_mcp modules (core/analysis/debug/memory/modify/stack/types/python) and kept synchronized in functionality, parameters, and return formats. However, project/session management tools (open_database, close_database) are implemented locally for per-project process orchestration. See the source in ida_domain_mcp/src/ida_domain_mcp/main.py and ida_domain_mcp/src/ida_domain_mcp/ida_tools.py for the bindings, and ida-pro-mcp/src/ida_pro_mcp/ida_mcp for upstream implementations.
Requirements
- Python: 3.12+
- IDA: IDA Pro 9.1.0 or later installed.
- uv is recommended for Python package and project management. See uv documentation for installation instructions.
Environment Variables
Configure the executable path according to ida-domain's documentation.
export IDADIR="[IDA Installation Directory]"
Headless executables (idat/idat64) must be available in the specified IDA installation directory.
Installation
Make sure to set up the environment variable as described above before running the MCP server.
Run with uvx
The simplest way to run the MCP server without installing anything is via uvx:
uvx ida-domain-mcp --transport http://127.0.0.1:8744
Install from PyPI
You can install the package as a dependency of your project from PyPI:
# With uv
uv init
uv add ida-domain-mcp
# Or with pip
pip install ida-domain-mcp
Running the MCP server
Two transport modes are supported by the server entrypoint ida-domain-mcp:
-
stdio (default, for direct MCP client integration)
uv run ida-domain-mcp --transport stdio -
SSE or Streamable HTTP (useful with the MCP Inspector and remote clients)
# sse at http://127.0.0.1:8744/sse uv run ida-domain-mcp --transport http://127.0.0.1:8744 # streamable http at http://127.0.0.1:8744/mcp uv run ida-domain-mcp --transport http://127.0.0.1:8744 --streamable-httpYou can then connect with the MCP Inspector for quick exploration:
npx @modelcontextprotocol/inspector # Point it to: http://127.0.0.1:8744/sse (SSE) or http://127.0.0.1:8744/mcp (Streamable HTTP)
Testing
Clone the repository and install the dependencies:
git clone https://github.com/xxyyue/ida_domain_mcp
cd ida_domain_mcp
uv sync
A simple dual-database test is provided:
# Start the server first (SSE mode)
uv run ida-domain-mcp --transport http://127.0.0.1:8744
# In another shell, run the test client
uv run tests/test_ida_mcp.py http://127.0.0.1:8744/sse
# Or, run the test agent
echo "OPENAI_API_KEY=sk-..." > .env
uv run tests/agent.py
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。